Blind Room Parameter Estimation Using Multiple Multichannel Speech Recordings

Srivastava, Prerak; Deleforge, Antoine

doi:10.1109/waspaa52581.2021.9632778

Cited by 10 publications

(8 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The task evaluated in this paper is similar to the one presented in [4]. Given a set of P two-channel noisy speech recordings made at different source-receiver positions in a room, the goal is to jointly estimate the room's volume V , total surface area S, and octave-band-wise reverberation times {RT 60 (b)} b∈B .…”

Section: Task Model and Training Setsmentioning

confidence: 99%

“…Contrary to [4], we do not estimate mean absorption coefficients here due to the lack of annotated real data for evaluation. Throughout this study, we use P = 3 speech recordings of 3 seconds each made with a 2-element linear array of aperture 22.5 cm placed parallel to the floor at 3 different positions and a fixed source position per room.…”

Section: Task Model and Training Setsmentioning

confidence: 99%

“…Throughout this study, we use P = 3 speech recordings of 3 seconds each made with a 2-element linear array of aperture 22.5 cm placed parallel to the floor at 3 different positions and a fixed source position per room. To perform the task, we use the neural network model in [4], which combines single-channel and inter-channel input features through several convolution, pooling and dense layers, and jointly estimates the means and variances of target quantities using a Gaussian maximum-likelihood loss. The input features are calculated using the short-time Fourier transform with a window size of 48 ms and 50% overlap.…”

Section: Task Model and Training Setsmentioning

confidence: 99%

“…Learning-based approaches have become ubiquitous in nearly all areas of audio signal processing. This includes the task of blind acoustic parameter estimation [1][2][3][4][5], which has applications in the automatic adaptation of augmented reality devices [6] or hearing aids [7] to their environment. A general bottleneck of learning-based methods is the need for large annotated training datasets.…”

Section: Introductionmentioning

confidence: 99%

“…We focus here on the task of jointly estimating the room's volume, total surface area and reverberation time per octave band using multiple, multichannel speech recordings, using our recently proposed state-of-the art model [4]. The model is tested on real reverberant speech signals from the dEchorate dataset [15] corresponding to 6 distinct microphone arrays, 7 distinct loudspeakers and 4 distinct room configurations.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Realistic Sources, Receivers and Walls Improve The Generalisability of Virtually-Supervised Blind Acoustic Parameter Estimators

Srivastava

Deleforge

2022

2022 International Workshop on Acoustic Signal Enhancement (IWAENC)

View full text Add to dashboard Cite

Blind acoustic parameter estimation consists in inferring the acoustic properties of an environment from recordings of unknown sound sources. Recent works in this area have utilized deep neural networks trained either partially or exclusively on simulated data, due to the limited availability of real annotated measurements. In this paper, we study whether a model purely trained using a fast image-source room impulse response simulator can generalize to real data. We present an ablation study on carefully crafted simulated training sets that account for different levels of realism in source, receiver and wall responses. The extent of realism is controlled by the sampling of wall absorption coefficients and by applying measured directivity patterns to microphones and sources. A state-of-the-art model trained on these datasets is evaluated on the task of jointly estimating the room's volume, total surface area, and octave-band reverberation times from multiple, multichannel speech recordings. Results reveal that every added layer of simulation realism at train time significantly improves the estimation of all quantities on real signals.

show abstract

Section: Task Model and Training Setsmentioning

confidence: 99%

Section: Task Model and Training Setsmentioning

confidence: 99%

Section: Task Model and Training Setsmentioning

confidence: 99%